[AMDGPU][MC] Add GFX12 VBUFFER encoding

Author: mbrkusanin

llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

@@ -373,7 +373,7 @@ class AMDGPUOperand : public MCParsedAsmOperand {
   bool isOffen() const { return isImmTy(ImmTyOffen); }
   bool isIdxen() const { return isImmTy(ImmTyIdxen); }
   bool isAddr64() const { return isImmTy(ImmTyAddr64); }
-  bool isOffset() const { return isImmTy(ImmTyOffset) && isUInt<16>(getImm()); }
+  bool isOffset() const { return isImmTy(ImmTyOffset); }
   bool isOffset0() const { return isImmTy(ImmTyOffset0) && isUInt<8>(getImm()); }
   bool isOffset1() const { return isImmTy(ImmTyOffset1) && isUInt<8>(getImm()); }
   bool isSMEMOffsetMod() const { return isImmTy(ImmTySMEMOffsetMod); }
@@ -1666,6 +1666,7 @@ class AMDGPUAsmParser : public MCTargetAsmParser {
   SMLoc getInstLoc(const OperandVector &Operands) const;
 
   bool validateInstruction(const MCInst &Inst, const SMLoc &IDLoc, const OperandVector &Operands);
+  bool validateOffset(const MCInst &Inst, const OperandVector &Operands);
   bool validateFlatOffset(const MCInst &Inst, const OperandVector &Operands);
   bool validateSMEMOffset(const MCInst &Inst, const OperandVector &Operands);
   bool validateSOPLiteral(const MCInst &Inst) const;
@@ -4143,6 +4144,40 @@ SMLoc AMDGPUAsmParser::getFlatOffsetLoc(const OperandVector &Operands) const {
   return getLoc();
 }
 
+bool AMDGPUAsmParser::validateOffset(const MCInst &Inst,
+                                     const OperandVector &Operands) {
+  auto Opcode = Inst.getOpcode();
+  auto OpNum = AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::offset);
+  if (OpNum == -1)
+    return true;
+
+  uint64_t TSFlags = MII.get(Inst.getOpcode()).TSFlags;
+  if ((TSFlags & SIInstrFlags::FLAT))
+    return validateFlatOffset(Inst, Operands);
+
+  if ((TSFlags & SIInstrFlags::SMRD))
+    return validateSMEMOffset(Inst, Operands);
+
+  const auto &Op = Inst.getOperand(OpNum);
+  if (isGFX12Plus() &&
+      (TSFlags & (SIInstrFlags::MUBUF | SIInstrFlags::MTBUF))) {
+    const unsigned OffsetSize = 24;
+    if (!isIntN(OffsetSize, Op.getImm())) {
+      Error(getFlatOffsetLoc(Operands),
+            Twine("expected a ") + Twine(OffsetSize) + "-bit signed offset");
+      return false;
+    }
+  } else {
+    const unsigned OffsetSize = 16;
+    if (!isUIntN(OffsetSize, Op.getImm())) {
+      Error(getFlatOffsetLoc(Operands),
+            Twine("expected a ") + Twine(OffsetSize) + "-bit unsigned offset");
+      return false;
+    }
+  }
+  return true;
+}
+
 bool AMDGPUAsmParser::validateFlatOffset(const MCInst &Inst,
                                          const OperandVector &Operands) {
   uint64_t TSFlags = MII.get(Inst.getOpcode()).TSFlags;
@@ -4800,10 +4835,7 @@ bool AMDGPUAsmParser::validateInstruction(const MCInst &Inst,
   if (!validateMovrels(Inst, Operands)) {
     return false;
   }
-  if (!validateFlatOffset(Inst, Operands)) {
-    return false;
-  }
-  if (!validateSMEMOffset(Inst, Operands)) {
+  if (!validateOffset(Inst, Operands)) {
     return false;
   }
   if (!validateMAIAccWrite(Inst, Operands)) {

llvm/lib/Target/AMDGPU/BUFInstructions.td

@@ -97,10 +97,17 @@ void AMDGPUInstPrinter::printNamedBit(const MCInst *MI, unsigned OpNo,
 void AMDGPUInstPrinter::printOffset(const MCInst *MI, unsigned OpNo,
                                     const MCSubtargetInfo &STI,
                                     raw_ostream &O) {
-  uint16_t Imm = MI->getOperand(OpNo).getImm();
+  uint32_t Imm = MI->getOperand(OpNo).getImm();
   if (Imm != 0) {
     O << " offset:";
-    printU16ImmDecOperand(MI, OpNo, O);
+
+    // GFX12 uses a 24-bit signed offset for VBUFFER.
+    const MCInstrDesc &Desc = MII.get(MI->getOpcode());
+    bool IsVBuffer = Desc.TSFlags & (SIInstrFlags::MUBUF | SIInstrFlags::MTBUF);
+    if (AMDGPU::isGFX12(STI) && IsVBuffer)
+      O << formatDec(SignExtend32<24>(Imm));
+    else
+      printU16ImmDecOperand(MI, OpNo, O);
   }
 }
 

llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp

@@ -2,11 +2,11 @@
 // RUN: not llvm-mc -triple=amdgcn -mcpu=tahiti %s 2>&1 | FileCheck --implicit-check-not=error: %s
 
 // offset too big
-// CHECK: :[[@LINE+1]]:{{[0-9]+}}: error: invalid operand for instruction
+// CHECK: :[[@LINE+1]]:{{[0-9]+}}: error: expected a 16-bit unsigned offset
 ds_add_u32 v2, v4 offset:1000000000
 
 // offset too big
-// CHECK: :[[@LINE+1]]:{{[0-9]+}}: error: invalid operand for instruction
+// CHECK: :[[@LINE+1]]:{{[0-9]+}}: error: expected a 16-bit unsigned offset
 ds_add_u32 v2, v4 offset:0x10000
 
 // offset0 twice

llvm/test/MC/AMDGPU/ds-err.s

@@ -1984,6 +1984,15 @@ s_cmp_nlt_f16 s1, s2
 s_singleuse_vdst 0x1234
 // CHECK: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
 
+buffer_atomic_sub_clamp_u32 v5, off, s[8:11], s3 offset:0 glc
+// CHECK: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_atomic_max_num_f32 v5, off, s[8:11], s3 offset:4095
+// CHECK: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_atomic_min_num_f32 v5, off, s[8:11], s3 offset:4095
+// CHECK: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
 ds_sub_clamp_rtn_u32 v5, v1, v2
 // CHECK: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU